docs: record maximum execution and local Docker tooling

This commit is contained in:
cesnimda
2026-11-18 13:15:35 +01:00
parent 271d37d1a8
commit 69b7ed2ea4
4 changed files with 372 additions and 0 deletions
+18
View File
@@ -1,5 +1,23 @@
# Direct execution continuation
## 18 November maximum execution and Windows Docker tooling
Verified start A31db5f488/roadmap271d37d1 against actualHEAD and remote. New maximum-backlog directive frozen in MAXIMUM_BACKLOG_EXECUTION_20261118.md. Three isolated workers plus root resumed; no duplicate landing recovery. Authoritative TODO/predecessor/owner gates remain active.
User requested a Windows rebuild/run script. Published/ref-readback A3 `dacdf16934b7634360b24d23a1d0356b1bcd4b3f` adds scripts/rebuild-windows-docker.ps1, standalone dev/windows-local.compose.yml, two offline configuration/PowerShell parser tests and docs/development/WINDOWS_DOCKER.md. Two tests and two ValidateOnly runs PASSED,0skips; credentials persist in ignored .env.windows-local and are not logged. Script builds the current checkout, local named-pipe Linux Docker only, projectjobjakt-windows-local, loopback3000, separate persistent MariaDB/files/tombstones, authenticated local admin. No productionCompose/.env, fetch/reset, imagepush or volume deletion. No AI/mail/provider workers. Actual Docker build/start NOT RUN: DesktopLinux engine pipe absent; CLI/Compose installed. User should start DockerDesktop then run the published script from the A3 checkout. Root AST-only graph update completed19,989nodes49,419edges; existing extractor warnings retained.
Owner asked whether local Ollama should be enabled. Recommended opt-in isolated local-only model/sidecar testing with synthetic data; this question is not explicit model-activation approval. No model/container activated or downloaded. Existing localmodel/recovery/security/provider gates preserved; localOllama alone does not qualify fullapplication.
Active packages (not yet accepted/integrated):
- PERF06 codex/analytics-query-batch from1db5: nativebaseline5commands active/4empty; candidate3/2 after combining onlyscalarcounts.51 native/related backend cases PASS; single additional Pomelo SQL-translation-only proof and immutable receipts/localcommit pending. No MariaDBexecution/load/throughput claim.
- A3 codex/a3-connected-followup from1db5: new2 ENdesktop/NBmobile native status→followup→calendar→clear journeys pass; fullaffected8 pass with21scans/captures. Legacy Norsk selector fixes preserve assertions. Serverlogs show checklist uniqueconstraint concurrencyerror despiteallbrowserassertionspassing; NOT closed, roottriage pending. Worker static/AST/evidence/localcommit pending; native ports5302/3300 freed.
- I18N00 B03 codex/i18n-message-census-b03 fromea18422a:346 source loci/28 backend owners,38 exactsourceblobs,10 associations and executableclassification/frontier QA inprogress. No runtimecatalogue/locale/statuspromotion; all24EU/all60 planned scope mandatory.
- Root codex/recovery-archive-validation from2fb10: offline archive verifier reproduced10 falseacceptance subtests (links/specialmembers, empty/wrongheader, unsafeDBpaths). New stdlib Python streaming validator checks regularmember/header without extraction. Final29 backupmethods and Bashsyntax pass; permissiontest fixture updated to actualgzip/header bytes after3 inheritedmockfixture failures. Firstdraftfixture Windows openconnection errors and correctedred retained separately; not applicationdefects. AST/evidence/review/publication pending. No realbackup/DBrestore/Docker/productionexecution; formatcheck is not SQLiteintegrity/recovery proof.
Root owns centralrecords, review/publishing/integration and nexteligible scheduling. No main/orchestrator/deployment. Preserve existing untracked graphs/screenshots/caches/originalproposals. Continue activepackages, then investigate logged checklist race before scheduling next independent work. Full1,665 suite is historicalsource-bound; do not repeat without changedrisk.
Directive SHA256 42800479bb2938dcacf2c3d6b95201a5e6561fc048b7f17e710d4599ab5f1952.
## 18 November public semantic accessibility qualification
A3/browser branches published and remotely read back `1db5f488f6138e4f0f33749452b7d9f47658f4b0`, application source36084473. Public header/main/footer landmarks now expose unchanged content correctly; loading has main as well. New regression FAILED on original missing-main source, final14 cases/three affected suites PASSED (overlaps13 from language correction), scoped lint/build/TypeScript/export/diff and AST-only graph19,898nodes49,311edges passed. Existing extractor warnings retained. Reused12-state browser matrix PASSED12 journeys and12 zero-violation axe scans with explicit assertions, preference PUTs remain zero. All12 before/after full-page PNG pairs byte-identical; selected images and hash pairs bound in docs/ui-modernisation/LANDING_LANDMARKS.md. No homepage visual redesign or content/action change; no native/human/whole-application accessibility acceptance. V27/route register updated as PARTIAL: static crawlability/locale metadata, older preview assets and design/claims gates remain open.
@@ -0,0 +1,333 @@
# Jobjakt — MAXIMUM BACKLOG EXECUTION
## Continuous, Parallel, Milestone-Driven Development
I want you to make substantially more progress through the entire Jobjakt development backlog during each execution.
**Do not treat a completed checkpoint, commit, test suite or 15-minute work cycle as a reason to stop.**
Continue for as long as the current execution environment, usage limits and available authorisation permit.
My objective is to complete as much of the ENTIRE outstanding roadmap as possible, not just a handful of small fixes.
## 1. Resume from the actual checkpoint
Latest reported remotely verified checkpoints:
- Application: `1db5f488`
- Roadmap: `271d37d1`
Reported verification:
- 1,665 frontend tests passed.
- 44 server-backed workflow tests passed.
- 12 rendered journeys and 12 accessibility scans passed.
First inspect:
- `docs/direct-roadmap/LATEST_CHECKPOINT.md`
- `docs/direct-roadmap/ROADMAP_TODO.md`
- `docs/direct-roadmap/ROADMAP_STATUS.md`
- Relevant source plans, acceptance documents and implementation records.
Verify the actual Git and worktree state.
Identify completed, partial, blocked and independently eligible work.
Preserve all previous work and accepted evidence.
## 2. Change execution strategy
Previously, development has often progressed through small individual fixes followed by checkpoint reports.
I now want a more ambitious execution strategy.
**Prioritise completing substantial backlog items and entire authorised milestones wherever feasible.**
Avoid repeatedly choosing tiny cosmetic or localisation fixes when larger independent functionality is ready.
Do not abandon necessary small fixes, particularly security or data-integrity issues.
Instead, group related changes into coherent implementation packages.
For example:
- Finish several related A3 workflow gaps together.
- Implement a meaningful portion of the European localisation infrastructure.
- Complete several independent performance optimisations.
- Resolve a group of verified security/dependency problems.
- Advance a complete Global Job Search architectural increment.
The goal is integrated, functional progress rather than simply increasing the commit count.
## 3. Execute across the complete roadmap
Work through ALL remaining authorised tasks, including:
### Phase 10
Complete independently executable acceptance evidence, preserve the 57 functionality groups and prepare remaining owner decisions.
### A3 modernisation
Finish remaining connected workflows, visual consistency, responsiveness, accessibility and error/recovery states.
### Global Job Search
Progress independently authorised country-aware search, geographic filtering, canonical location models, occupation data and provider-independent functionality.
Respect Phase 10, Product Review, initiative approvals and provider-rights gates. Do not treat previous planning approval as blanket permission for protected implementation.
### European localisation
Move beyond EN/NB maintenance.
The objective remains all 24 official EU languages plus the wider approved European scope.
Progress locale infrastructure, translation catalogues, backend localisation, documents, discovery, the homepage, validation tooling and language coverage.
Do not claim locales are complete merely because they have been researched or registered.
### Performance
Complete as many of the 18 optimisation areas as justified by evidence.
Prioritise the next eligible performance tasks identified in the continuation record.
### Public homepage
Advance authorised design-neutral technical work and eligible implementation, while preserving any outstanding design approval.
### Security
Triage and remediate verified issues within authorised scope.
Continue independent security preparation and safe offline checks.
### Release readiness
Progress isolated qualification, reproducible builds, operational diagnostics and recovery evidence.
Do not bypass production or formal security gates.
### Product Review and technical debt
Complete independent preparatory work where allowed. Respect formal predecessor approvals.
## 4. Use maximum safe parallelism
Where available, use multiple Codex workers or sub-agents.
I would prefer approximately 3–4 concurrent independent workstreams when the available environment and resources support them.
Possible assignments:
**Worker A — Frontend/A3**
- Connected workflows.
- Visual discrepancies.
- Accessibility.
- UI performance.
**Worker B — European I18N/Global Jobs**
- Locale architecture.
- Translation tooling.
- Geographic infrastructure.
- Provider-independent discovery.
**Worker C — Backend/Performance**
- Database queries.
- Caching.
- Pagination.
- API efficiency.
- Safe concurrency fixes.
**Worker D — Security/Operations**
- Dependency remediation.
- Offline qualification.
- Backup/recovery evidence.
- Operational diagnostics.
These are suggested allocations, not fixed assignments.
Choose tasks according to actual eligibility and dependencies.
Use isolated worktrees or nonconflicting file ownership.
Avoid simultaneous changes to shared mutable databases, schema migrations and common configuration files.
Do not overload the host with unnecessary concurrent builds or tests.
If fewer workers are available, continue sequentially.
**Parallelism must improve completed, validated work—not merely produce more branches.**
## 5. Avoid unnecessary repetitive verification
The recent execution ran the full 1,665 frontend tests.
That is valuable acceptance evidence, but I do not want the entire suite unnecessarily rerun after every small, isolated change.
Adopt risk-based verification:
- Run targeted tests during development.
- Run affected integration tests for connected changes.
- Run typecheck/lint/build as appropriate.
- Run broader regression suites when required by affected scope, integration or authoritative acceptance gates.
- Reuse valid evidence only when the source and test conditions remain applicable.
Never weaken tests or skip mandatory acceptance requirements.
Use the saved execution time for further implementation.
## 6. Continuous execution loop
After completing an implementation package:
1. Verify the work.
2. Fix introduced regressions.
3. Commit validated changes.
4. Publish when available and authorised.
5. Update the authoritative TODO and continuation checkpoint.
6. Recalculate the eligible backlog.
7. Assign the next independent tasks.
8. **Immediately continue implementing them.**
Repeat this cycle.
Do not voluntarily end execution merely because a coherent package has been completed.
Do not ask whether I want you to continue.
Do not produce a final handoff after each package unless the environment forces the execution to end.
## 7. Dependency-aware scheduling
Treat the roadmap as a dependency graph.
Prioritise work that:
- Removes blockers for other milestones.
- Improves central user journeys.
- Resolves confirmed security or integrity defects.
- Completes partially implemented functionality.
- Enables European language and market expansion.
- Provides measurable performance improvements.
If a task is blocked:
- Record the blocker.
- Identify the required decision or prerequisite.
- Preserve current progress.
- Move to another eligible task.
Do not repeatedly retry the same unchanged external failure.
Do not sit idle waiting for approval while independent work remains.
## 8. Minimise interruptions
I want minimal human intervention.
Do not ask questions about routine implementation choices already covered by approved plans.
Maintain a consolidated owner-decision register.
When a protected decision is needed:
- Record the question.
- Explain the options.
- Recommend an approach.
- Identify blocked work.
- Continue other eligible tasks.
Only present questions when no further authorised independent work remains, except where immediate security or data-loss risks require attention.
Batch questions wherever possible.
Do not self-approve protected gates.
## 9. Preserve important requirements
All existing application requirements remain active.
In particular:
- Entirely European product architecture.
- All 24 EU official languages and broader European scope.
- Country-aware maps, location filtering and job providers.
- Clean A3 visual design on desktop and mobile.
- No excessive nested cards.
- Full preservation of existing features.
- Correct tenant/user isolation.
- Secure authentication and MFA.
- CV revision and document integrity.
- Existing PDF acceptance requirements.
- Correct recovery and backup behaviour.
- Accessible user journeys.
Do not optimise away features or compromise correctness.
## 10. Manage execution limits
Use the current session's available execution capacity productively.
Do not artificially stop after a short development interval.
However, do not attempt to bypass platform usage limits, context limits or execution restrictions.
If approaching an unavoidable limit, prioritise preserving recoverable work:
- Commit complete validated changes.
- Preserve unfinished files and worktrees.
- Record incomplete tests.
- Update the continuation checkpoint.
- Identify the exact next eligible task.
- Document any uncommitted or unpublished work.
Do not claim work will automatically continue after a session ends unless an actual supported continuation mechanism is configured.
## 11. Meaningful progress reporting
Track cumulative progress across the whole roadmap.
For each workstream, report:
- Outstanding tasks at the start.
- Tasks completed.
- Tasks partially completed.
- Tasks blocked.
- Actual implemented functionality.
- Tests and acceptance evidence.
- Commits and publication status.
- Next eligible work.
Focus on meaningful milestone completion, not the number of commits or files changed.
Do not claim full acceptance without the corresponding evidence and owner decisions.
## 12. Non-negotiable restrictions
- Do not use or reactivate the autonomous orchestrator.
- Do not merge into `main`.
- Do not deploy production.
- Do not perform destructive migrations without approval.
- Do not bypass mandatory acceptance or security gates.
- Do not activate unqualified AI models or unapproved providers.
- Do not incur external costs without approval.
- Do not discard uncommitted work.
- Do not fabricate verification results.
## FINAL DIRECTIVE
**START MAXIMUM BACKLOG EXECUTION NOW.**
I want as much of my entire Jobjakt roadmap implemented as possible.
Use substantial implementation packages rather than unnecessarily small checkpoints.
Run independent workstreams concurrently where safe.
Use targeted verification during development and broader tests at appropriate integration gates.
After every completed package, immediately begin another eligible task.
If one task is blocked, move to another.
Continue until:
- No further independently authorised useful work remains; or
- An unavoidable technical or usage limit prevents continuation.
Only then provide a consolidated development report and outstanding owner decisions.
**Do not voluntarily finish after 15 minutes or after completing only a few tasks. Keep progressing for as long as you can safely and legitimately execute.**
+18
View File
@@ -1,5 +1,23 @@
# Direct roadmap status
## 18 November maximum execution and Windows Docker tooling
Verified start A31db5f488/roadmap271d37d1 against actualHEAD and remote. New maximum-backlog directive frozen in MAXIMUM_BACKLOG_EXECUTION_20261118.md. Three isolated workers plus root resumed; no duplicate landing recovery. Authoritative TODO/predecessor/owner gates remain active.
User requested a Windows rebuild/run script. Published/ref-readback A3 `dacdf16934b7634360b24d23a1d0356b1bcd4b3f` adds scripts/rebuild-windows-docker.ps1, standalone dev/windows-local.compose.yml, two offline configuration/PowerShell parser tests and docs/development/WINDOWS_DOCKER.md. Two tests and two ValidateOnly runs PASSED,0skips; credentials persist in ignored .env.windows-local and are not logged. Script builds the current checkout, local named-pipe Linux Docker only, projectjobjakt-windows-local, loopback3000, separate persistent MariaDB/files/tombstones, authenticated local admin. No productionCompose/.env, fetch/reset, imagepush or volume deletion. No AI/mail/provider workers. Actual Docker build/start NOT RUN: DesktopLinux engine pipe absent; CLI/Compose installed. User should start DockerDesktop then run the published script from the A3 checkout. Root AST-only graph update completed19,989nodes49,419edges; existing extractor warnings retained.
Owner asked whether local Ollama should be enabled. Recommended opt-in isolated local-only model/sidecar testing with synthetic data; this question is not explicit model-activation approval. No model/container activated or downloaded. Existing localmodel/recovery/security/provider gates preserved; localOllama alone does not qualify fullapplication.
Active packages (not yet accepted/integrated):
- PERF06 codex/analytics-query-batch from1db5: nativebaseline5commands active/4empty; candidate3/2 after combining onlyscalarcounts.51 native/related backend cases PASS; single additional Pomelo SQL-translation-only proof and immutable receipts/localcommit pending. No MariaDBexecution/load/throughput claim.
- A3 codex/a3-connected-followup from1db5: new2 ENdesktop/NBmobile native status→followup→calendar→clear journeys pass; fullaffected8 pass with21scans/captures. Legacy Norsk selector fixes preserve assertions. Serverlogs show checklist uniqueconstraint concurrencyerror despiteallbrowserassertionspassing; NOT closed, roottriage pending. Worker static/AST/evidence/localcommit pending; native ports5302/3300 freed.
- I18N00 B03 codex/i18n-message-census-b03 fromea18422a:346 source loci/28 backend owners,38 exactsourceblobs,10 associations and executableclassification/frontier QA inprogress. No runtimecatalogue/locale/statuspromotion; all24EU/all60 planned scope mandatory.
- Root codex/recovery-archive-validation from2fb10: offline archive verifier reproduced10 falseacceptance subtests (links/specialmembers, empty/wrongheader, unsafeDBpaths). New stdlib Python streaming validator checks regularmember/header without extraction. Final29 backupmethods and Bashsyntax pass; permissiontest fixture updated to actualgzip/header bytes after3 inheritedmockfixture failures. Firstdraftfixture Windows openconnection errors and correctedred retained separately; not applicationdefects. AST/evidence/review/publication pending. No realbackup/DBrestore/Docker/productionexecution; formatcheck is not SQLiteintegrity/recovery proof.
Root owns centralrecords, review/publishing/integration and nexteligible scheduling. No main/orchestrator/deployment. Preserve existing untracked graphs/screenshots/caches/originalproposals. Continue activepackages, then investigate logged checklist race before scheduling next independent work. Full1,665 suite is historicalsource-bound; do not repeat without changedrisk.
Directive SHA256 42800479bb2938dcacf2c3d6b95201a5e6561fc048b7f17e710d4599ab5f1952.
## 18 November public semantic accessibility qualification
A3/browser branches published and remotely read back `1db5f488f6138e4f0f33749452b7d9f47658f4b0`, application source36084473. Public header/main/footer landmarks now expose unchanged content correctly; loading has main as well. New regression FAILED on original missing-main source, final14 cases/three affected suites PASSED (overlaps13 from language correction), scoped lint/build/TypeScript/export/diff and AST-only graph19,898nodes49,311edges passed. Existing extractor warnings retained. Reused12-state browser matrix PASSED12 journeys and12 zero-violation axe scans with explicit assertions, preference PUTs remain zero. All12 before/after full-page PNG pairs byte-identical; selected images and hash pairs bound in docs/ui-modernisation/LANDING_LANDMARKS.md. No homepage visual redesign or content/action change; no native/human/whole-application accessibility acceptance. V27/route register updated as PARTIAL: static crawlability/locale metadata, older preview assets and design/claims gates remain open.
+3
View File
@@ -4,6 +4,9 @@ Updated 18 November 2026. This is the cross-phase work list. The source prompts,
- [x] **Public semantic landmarks, bounded accessibility fix**.36084473/evidence1db5f488 published/ref-readback;14 affected cases,12 controlled journeys/12 scoped zero-violation axe scans and12 byte-identical screenshot pairs; lint/build/TypeScript/AST passed. V27 partial: homepage SEO/assets/design/native acceptance open.
- [x] **Windows local Docker rebuild/run tooling**.dacdf169 published/ref-readback;2 offline configuration/parser cases and2 ValidateOnly calls passed. Persistent isolated localproject, loopback/auth, no model/provider workers. Actualimagebuild/start NOTRUN: localDesktopLinuxengine unavailable. UserOllama question remains recommendation, noactivation approval.
- [ ] **Current coherent packages**.PERF06 analytics, connectedfollowup/calendar and I18N00B03 are active in isolatedworktrees; rootarchiveformatfix29methods passed but publicationpending. Checklistuniqueconstraint log remains open despite8 browsertests passing. Seecheckpoint before claiming completion.
## 18 November validated increments and active qualification
- [x] **Scoped Jest glob/minimatch advisory remediation, Windows Node24 qualification**.418de0ad integrated0765b151, published/ref-readback. Full129 suites/1,665 tests and separate3 actual-Jest-caller checks passed, locked install/lint/typecheck/build/AST passed. Root current E install/full lint/build passed.48 high propagated audit records represent ONE remaining picomatch advisory; no blanket security clearance. Node20/Linux/hosted open.